Run this workflow on InstaSD
Get started in minutes! Run this ComfyUI workflow online - no setup required.
Description
Leveraging SageAttention, TeaCache, and torch.compile for significantly improved speed and efficiency.
š Benchmark Results
Hardware: RTX 4090 | Settings: 64 Frames | 16 FPS | 20 Steps
ā±ļø Base: 1002s
ā” Optimized (SageAttention + TeaCache + torch.compile): 516s
š„ Twice as fast!
āØ Key Features
ā Drastically Faster Inference ā Cut generation time in half!
ā SageAttention Optimization ā More efficient attention mechanisms
ā TeaCache Acceleration ā Reduces redundant computations
ā torch.compile ā Compiler optimizations for speedup
Models
| File | Destination | Source |
|---|---|---|
| umt5_xxl_fp8_e4m3fn_scaled.safetensors | /ComfyUI/models/text_encoders | Download |
| wan_2.1_vae.safetensors | /ComfyUI/models/vae | Download |
| clip_vision_h.safetensors | /ComfyUI/models/clip_vision | Download |
| wan2.1_i2v_720p_14B_bf16.safetensors | /ComfyUI/models/diffusion_models | Download |
| burger.jpg | /ComfyUI/input | Download |
Nodes
CLIPLoaderCLIPTextEncodeVAELoaderVAEDecodeTiledNoteCLIPVisionLoaderCLIPVisionEncodeLoadImageUNETLoaderTorchCompileModelWanVideoVHS_VideoCombineWanImageToVideoPathchSageAttentionKJWanVideoTeaCacheKJKSampler